For over a decade I've been turning business processes into systems that keep working after everyone stops watching: audits, architecture, applications, integrations, the infrastructure underneath and the reporting on top. Most of that work sits behind NDAs, so what you'll find here isn't the portfolio — it's the method, and the things I can publish.
What changed recently is how it gets built. I now develop with AI as the main instrument rather than as autocomplete — which turns out to be a discipline, not a shortcut. It works when the specification is a real contract, when decisions are written down with their reasoning, and when every claim is checked against a running deployment instead of a green test. It fails, quietly and expensively, when any of the three is missing. Most of my current work is about getting that right at production scale.
An AI-native publishing network where autonomous agents and humans publish, read, cite and challenge each other through open APIs. It's the project where I take everything to the edge of what the platform allows, on purpose.
- Every article carries checkable provenance — who wrote it, which agent published it, which human answers for it, and a signature over the revision that gets verified.
- Agents are first-class readers and authors: a REST API and an MCP endpoint, tokens, keys, quotas, the same rules people get.
- One runtime for all of it — Cloudflare Workers, D1, R2, Queues, Durable Objects, Analytics Engine and Workers AI. No servers, no containers.
- Built spec-first: a specification that decides arguments, ADRs for every reversal, boundary rules that fail the build, and checkpoint scripts that run against a live deployment on every push — because the last several real bugs were the kind no unit test can see.
- 🤖 Automation that runs a business, not a demo. Workflows, integrations and AI agents on n8n — API integrations, marketplace pipelines, knowledge-base agents, support bots, custom-trained models.
- 📱 Applications people actually work in. Internal tools on AppSheet and BigQuery, from inventory and asset tracking to custom CRM and an HR recruiting platform — delivered in weeks, not quarters.
- 📊 Data and analytics end to end. Schema design and optimisation on PostgreSQL, MySQL, BigQuery and ClickHouse; live dashboards in Looker Studio, Metabase and Power BI with KPI, budget and ROI actually visible.
- 🐳 Infrastructure I run myself. Self-hosted Docker stacks with persistent data, backups and management scripts — n8n, Caddy, databases, identity (Authentik, Authelia), monitoring.
- ☁️ Edge and web. Public APIs and static tools on Cloudflare Workers and Pages, in TypeScript and JavaScript.
Client work is mostly under NDA. If a case sounds like yours, ask me about it.
Self-hosted stacks — Docker Compose, persistent data, backup scripts, ready for production.
| n8n ⭐12 | WordPress ⭐7 | Authentik ⭐4 |
| Beszel ⭐4 | Caddy ⭐3 | Gatus ⭐3 |
| Qdrant ⭐3 | ClickHouse | Ollama + Open WebUI |
| Prometheus + Grafana | Metabase | PostgreSQL · MySQL · Redis |
APIs on Cloudflare Workers — one job each, JSON / text / YAML / XML, no key required.
whois · dns · ip · meta · uuid · translit · num2words
Browser-only tools — static and private; nothing leaves the tab.
md ⭐5 — markdown live preview · pass ⭐3 — passwords and passphrases · json — JSON toolkit · random — text randomiser · uuid · ulid · utm
Offline-first — mytasks, a Flutter task manager with no accounts, no analytics and no network calls.
- Website: airat.top — services, case list and contact
- Blog: blog.airat.top
- Email: mail@airat.top
- Status: status.airat.top
